Fast Thomas Process (Rcpp-backed) in an arbitrary polygon

Simulate \(n\_target\) points from a Thomas (Gaussian Neyman–Scott) process, using a fast C++ generator in the domain's bounding box and then filtering to the polygon with sf. This is a drop-in replacement for the spatstat-based Thomas simulator and is substantially faster for large simulations.

Usage

rthomas_fast(
  domain,
  n_target,
  kappa = NULL,
  mu = 10,
  sigma = 1,
  oversample = 1.3
)

Arguments

domain

An sf polygon/multipolygon defining the study area (projected CRS recommended).

n_target

Integer, desired number of retained points inside domain.

kappa

Optional parent intensity (parents per unit area). If NULL, it is derived from n_target, mu, and domain area.

mu

Mean offspring per parent (Poisson).

sigma

Gaussian cluster scale (standard deviation of offspring displacement; map units).

oversample

Scalar \(> 1\). Multiplier used to request enough raw offspring in the bbox so that, after polygon filtering, you still retain about n_target. Default 1.3.

Value

An sf POINT layer with exactly n_target rows. (If the Thomas generator yields too few points, the function generates additional raw points and clips again; as a final fallback it draws uniform points in the polygon.)

Details

Let \(A_D\) be the polygon area and \(A_B\) its bbox area. The expected fraction of bbox points that survive the polygon filter is \(f \approx A_D / A_B\). We request roughly n_target / f points from C++ (times oversample), then filter, and finally thin to return exactly n_target. If clipping leaves too few points, the function generates additional raw bbox points and clips again (never by duplicating existing points). As a final fallback it uses uniform sampling inside the polygon to fill any remaining shortfall.

Examples

if (FALSE) { # \dontrun{
dom <- create_sampling_domain()
pts <- rthomas_fast(dom, n_target = 2000, mu = 10, sigma = 1)
plot(sf::st_geometry(pts), pch = 16, cex = 0.4)
} # }